nemo-platform

NeMo Platform is an open-source platform from NVIDIA that consolidates the NeMo libraries into a single command-line interface, Python SDK, and web UI. It targets teams shipping production AI agents, with a focus on making them faster, more accurate, and safer. The platform provides several capability areas. For secure agents, it includes guardrails for content safety, jailbreak detection, and PII redaction, plus an Auditor for red-teaming and an Anonymizer for training data. For evaluation, it supports LLM-as-judge, deterministic, agentic, and RAG benchmarks, with Harbor-backed eval suites for regression testing. Tuning features cover skill optimization, prompt and hyperparameter tuning, and Switchyard model routing. Agent building uses the NVIDIA NeMo Agent Toolkit for LangGraph-based agents, backed by shared infrastructure such as an Inference Gateway, Secrets, Files, Entity Store, and Jobs. A Data Designer component generates synthetic data for training or evaluation. Installation is available from PyPI using uv, with a global `nemo` command. A source checkout is supported for development, with Flox recommended for toolchain management. The `nemo setup` command starts local services, registers an LLM provider, discovers models, installs agent skills, and can deploy a sample calculator agent. The platform also exposes REST APIs, including an OpenAI-compatible chat completions endpoint through the inference gateway. NeMo Studio, a browser UI for chat, monitoring, and optimization suggestions, is included as an alpha component. The CLI remains the primary interface. Skills can be installed into coding agents such as Claude Code, Cursor, Codex, and OpenCode, enabling tasks like scaffolding agents, running evaluations, adding guardrails, and optimizing deployments. The project is licensed under Apache 2.0 and supports Python 3.12-3.13. Documentation covers setup, CLI and API references, and telemetry and privacy controls.